Skip to main content
ARS Home » Research » Publications at this Location » Publication #143278


item Lawrence, Kurt
item Windham, William
item Park, Bosoon
item Buhr, Richard - Jeff

Submitted to: Near Infrared Spectroscopy Journal
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/10/2003
Publication Date: 10/27/2003
Citation: Lawrence, K.C., Windham, W.R., Park, B., Buhr, R.J. 2003. A hyperspectral imaging system for identification of fecal and ingesta contamination on poultry carcasses. Journal Near Infrared Spectroscopy. 11(4):269-281.

Interpretive Summary: In a poultry processing plant, a potential source of pathogen contamination on a poultry carcass is poultry feces and ingesta. Food safety regulations require that inspector check to ensure there are no contaminants on the carcasses before they enter the chiller tanks. Current methods require the inspectors to remove at least 10 birds per hour to check for contaminants. However, since carcasses travel on processing-line shackles at a speed of about 140 birds per minute, 8400 carcasses are processed per hour. We are developing a research imaging system to check for fecal or ingesta (undigested food) contamination on carcasses. The system collects successive images at many different wavelengths which can be used to optimize the contrast between the contaminant and the rest of the carcasses. This paper describes the research imaging system, known as a hyperspectral imaging system and its calibration. Results that identify key wavelengths in the visible light range for detecting contaminants are presented and these key-wavelengths were then tested on images of whole carcasses contaminated with small spots of feces and ingesta, which were collected with the hyperspectral imaging system. A detection algorithm was also developed that enhances the separation between the carcass and the contaminants. Results show more than 96% of the contaminant spots were detected. More work is needed to see if the chicken's diet and other processing variables, such as scald-water temperature, affect the results. These wavelengths and detection algorithms will later be applied in a multispectral imaging camera with special filters so that a final system can operate at processing line speeds.

Technical Abstract: A method and system for detecting fecal and ingesta contaminates were demonstrated. A visible / near infrared monochromator, which measured reflectance, and principal component analysis were first used to identify key wavelengths from fecal and uncontaminated skin samples. Measurements at 434, 517, 565, and 628 nm were identified and used for evaluation with a hyperspectral imaging system. The hyperspectral imaging system, which was a line-scan (pushbroom) imaging system, consisted of a hyperspectral camera, fiber-optic line lights, a computer, and frame grabber. The hyperspectral imaging camera consisted of a high resolution CCD camera, a prism-grating-prism spectrograph, focusing lens, associated optical hardware, and a motorized controller. The imaging system operated from about 400 to 900 nm. The hyperspectral imaging system was calibrated for wavelength, distance, and percent reflectance, and analysis of calibrated images at the key wavelengths indicated that single-wavelength images were inadequate for detecting contaminates. However, a ratio of images at two of the key wavelengths was able to identify fecal and ingesta contaminates. Specifically, the ratio of the 565-nm image divided by the 517-nm image produced good results. The ratio image was then further processed by masking the background and either enhancing the image contrast with a nonlinear-histogram stretch, or applying a fecal threshold. The results indicated that, for the limited sample population, more than 96% of the contaminants were detected. Thus, the hyperspectral imaging system was able to detect contaminants and showed feasibility, but was too slow for real-time on-line processing. Therefore, a multivariate system operating at 565 and 517 nm, which should be capable of operating at real-time on-line processing speed, should be used. Further research with such a system needs to be conducted.